[D] Add highlighting for recognised __traits/pragmas

[ichordev]

Changes:

• Valid __traits are highlighted with constant.other.d.
• Invalid __traits now trigger invalid.illegal.d, because they cause compile errors.
• Highlight known pragmas with constant.other.d. Invalid pragmas are not meant to cause compile errors in D (currently they do, but it's being fixed), so they do NOT trigger invalid.illegal.d.

[deathaxe]

A certain identifier not being supported by a compiler (causing compiler errors) may not directly be equal to invalid syntax.

In general being too over-eager may be annyoing to end users due to seeeing tons of red (illegal highlighted) code all the time while editing code.

That's however too late to question for D anyway though.

[ichordev]

A certain identifier not being supported by a compiler (causing compiler errors) may not directly be equal to invalid syntax.

Valid __traits are specified in D's grammar. The reference compiler (DMD) gives you an error, so LDC and GDC will both give you an error too because they use the same backend. As far as I know, this is intentional behaviour.

Like I noted, unrecognised pragmas are not supposed to cause errors because implementation-specific pragmas are provided for by the spec, and so they don't trigger invalid syntax highlighting here. Instead, unrecognised pragmas just don't get special highlighting so that you can see that something is wrong (i.e. you probably made a typo).

[deathaxe]

A synatx definition should highlight tokens and ideally handle incomplete token/statements in sane ways to create the impression of it already knowing what's up next.

It is not necessarily required to do a linter's job. Even though the compiler isn't happy with a certain token, currently, support for new ones may be added at any time. So unsupported pragmas are not a syntax error. The compiler's parser would also read the full token, but just throw an error at a later point, as it doesn't know it.

[ichordev]

So unsupported pragmas are not a syntax error. The compiler's parser would also read the full token, but just throw an error at a later point, as it doesn't know it.

Yes, pragmas are designed to accommodate compiler-specific identifiers, so my PR already accounts for the fact that their identifiers are arbitrary by just ignoring unrecognised identifiers.

Traits, on the other hand, are defined as an exhaustive list of keywords. A traits expression, by definition, must use one of these keywords. Yes, new keywords may be added, but the same holds for every other aspect of the language's syntax.